Inferring fruit infestation prevalence from a combination of pre-harvest monitoring and consignment sampling data

TL;DR

This paper introduces a method to estimate fruit infestation rates by combining pre-harvest monitoring and consignment sampling data using a Bayesian model.

Contribution

A hierarchical Bayesian model is developed to integrate in-field and consignment data for infestation prevalence estimation.

Findings

Pre-harvest monitoring with sufficient trap density can provide 95% belief in very low infestation prevalence.

High inspection sensitivity and multiple consignment samples can confirm low infestation rates without pre-harvest data.

The model accounts for uncertainty in inspection and monitoring methods.

Abstract

International trade in horticultural produce happens under phytosanitary inspection and production protocols. Fruit inspection typically involves the sampling and inspection of either 600-pieces or 2% of packed product within a single consignment destined for export, with the purpose of certification (typically with 95% confidence) that the true infestation level within the consignment in question doesn’t exceed a pre-specified design prevalence. Sampling of multiple consignments from multiple production blocks in conjunction with pre-harvest monitoring for pests can be used to provide additional inference on the prevalence of infested fruit within an overall production system subject to similar protocols. Here we develop a hierarchical Bayesian model that combines in-field monitoring data with consignment sample inspection data to infer the prevalence of infested fruit in a production system. The results illustrate how infestation prevalence is influenced by the number of consignments inspected, the detection efficacy of consignment sampling, and in-field monitoring effort and sensitivity. Uncertainty in inspection performance, monitoring methods, and exposure of fruit to pests is accommodated using statistical priors within a Bayesian modelling framework. We demonstrate that pre-harvest surveillance with a sufficient density of traps and moderate detection sensitivity can provide 95% belief that the prevalence of infestation is below \documentclass[12pt]{minimal}
\usepackage{amsmath}
\usepackage{wasysym}
\usepackage{amsfonts}
\usepackage{amssymb}
\usepackage{amsbsy}
\usepackage{mathrsfs}
\usepackage{upgreek}
\setlength{\oddsidemargin}{-69pt}
\begin{document}$$1 \times 10^{-6}$$\end{document}1×10-6. In the absence of pre-harvest monitoring, it is still possible to gain high confidence in a very low prevalence of infestation (\documentclass[12pt]{minimal}
\usepackage{amsmath}
\usepackage{wasysym}
\usepackage{amsfonts}
\usepackage{amssymb}
\usepackage{amsbsy}
\usepackage{mathrsfs}
\usepackage{upgreek}
\setlength{\oddsidemargin}{-69pt}
\begin{document}$$<1 \times 10^{-5}$$\end{document}<1×10-5) on the basis of multiple clean samples if the inspection sensitivity during consignment sampling is high and sufficient consignments are inspected. Our work illustrates the cumulative power of in-field surveillance and consignment sampling to update estimates of infestation prevalence.

Figures